1 /*
2  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Voltaire, Inc.  All rights reserved.
4  * Copyright (c) 2006 Intel Corporation.  All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer in the documentation and/or other materials
23  *        provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 
35 #include <linux/module.h>
36 #include <linux/init.h>
37 #include <linux/err.h>
38 #include <linux/random.h>
39 #include <linux/spinlock.h>
40 #include <linux/slab.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/kref.h>
43 #include <linux/idr.h>
44 #include <linux/workqueue.h>
45 #include <uapi/linux/if_ether.h>
46 #include <rdma/ib_pack.h>
47 #include <rdma/ib_cache.h>
48 #include <rdma/rdma_netlink.h>
49 #include <net/netlink.h>
50 #include <uapi/rdma/ib_user_sa.h>
51 #include <rdma/ib_marshall.h>
52 #include <rdma/ib_addr.h>
53 #include <rdma/opa_addr.h>
54 #include "sa.h"
55 #include "core_priv.h"
56 
57 #define IB_SA_LOCAL_SVC_TIMEOUT_MIN		100
58 #define IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT		2000
59 #define IB_SA_LOCAL_SVC_TIMEOUT_MAX		200000
60 #define IB_SA_CPI_MAX_RETRY_CNT			3
61 #define IB_SA_CPI_RETRY_WAIT			1000 /*msecs */
62 static int sa_local_svc_timeout_ms = IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT;
63 
64 struct ib_sa_sm_ah {
65 	struct ib_ah        *ah;
66 	struct kref          ref;
67 	u16		     pkey_index;
68 	u8		     src_path_mask;
69 };
70 
71 enum rdma_class_port_info_type {
72 	RDMA_CLASS_PORT_INFO_IB,
73 	RDMA_CLASS_PORT_INFO_OPA
74 };
75 
76 struct rdma_class_port_info {
77 	enum rdma_class_port_info_type type;
78 	union {
79 		struct ib_class_port_info ib;
80 		struct opa_class_port_info opa;
81 	};
82 };
83 
84 struct ib_sa_classport_cache {
85 	bool valid;
86 	int retry_cnt;
87 	struct rdma_class_port_info data;
88 };
89 
90 struct ib_sa_port {
91 	struct ib_mad_agent *agent;
92 	struct ib_sa_sm_ah  *sm_ah;
93 	struct work_struct   update_task;
94 	struct ib_sa_classport_cache classport_info;
95 	struct delayed_work ib_cpi_work;
96 	spinlock_t                   classport_lock; /* protects class port info set */
97 	spinlock_t           ah_lock;
98 	u8                   port_num;
99 };
100 
101 struct ib_sa_device {
102 	int                     start_port, end_port;
103 	struct ib_event_handler event_handler;
104 	struct ib_sa_port port[0];
105 };
106 
107 struct ib_sa_query {
108 	void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
109 	void (*release)(struct ib_sa_query *);
110 	struct ib_sa_client    *client;
111 	struct ib_sa_port      *port;
112 	struct ib_mad_send_buf *mad_buf;
113 	struct ib_sa_sm_ah     *sm_ah;
114 	int			id;
115 	u32			flags;
116 	struct list_head	list; /* Local svc request list */
117 	u32			seq; /* Local svc request sequence number */
118 	unsigned long		timeout; /* Local svc timeout */
119 	u8			path_use; /* How will the pathrecord be used */
120 };
121 
122 #define IB_SA_ENABLE_LOCAL_SERVICE	0x00000001
123 #define IB_SA_CANCEL			0x00000002
124 #define IB_SA_QUERY_OPA			0x00000004
125 
126 struct ib_sa_service_query {
127 	void (*callback)(int, struct ib_sa_service_rec *, void *);
128 	void *context;
129 	struct ib_sa_query sa_query;
130 };
131 
132 struct ib_sa_path_query {
133 	void (*callback)(int, struct sa_path_rec *, void *);
134 	void *context;
135 	struct ib_sa_query sa_query;
136 	struct sa_path_rec *conv_pr;
137 };
138 
139 struct ib_sa_guidinfo_query {
140 	void (*callback)(int, struct ib_sa_guidinfo_rec *, void *);
141 	void *context;
142 	struct ib_sa_query sa_query;
143 };
144 
145 struct ib_sa_classport_info_query {
146 	void (*callback)(void *);
147 	void *context;
148 	struct ib_sa_query sa_query;
149 };
150 
151 struct ib_sa_mcmember_query {
152 	void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
153 	void *context;
154 	struct ib_sa_query sa_query;
155 };
156 
157 static LIST_HEAD(ib_nl_request_list);
158 static DEFINE_SPINLOCK(ib_nl_request_lock);
159 static atomic_t ib_nl_sa_request_seq;
160 static struct workqueue_struct *ib_nl_wq;
161 static struct delayed_work ib_nl_timed_work;
162 static const struct nla_policy ib_nl_policy[LS_NLA_TYPE_MAX] = {
163 	[LS_NLA_TYPE_PATH_RECORD]	= {.type = NLA_BINARY,
164 		.len = sizeof(struct ib_path_rec_data)},
165 	[LS_NLA_TYPE_TIMEOUT]		= {.type = NLA_U32},
166 	[LS_NLA_TYPE_SERVICE_ID]	= {.type = NLA_U64},
167 	[LS_NLA_TYPE_DGID]		= {.type = NLA_BINARY,
168 		.len = sizeof(struct rdma_nla_ls_gid)},
169 	[LS_NLA_TYPE_SGID]		= {.type = NLA_BINARY,
170 		.len = sizeof(struct rdma_nla_ls_gid)},
171 	[LS_NLA_TYPE_TCLASS]		= {.type = NLA_U8},
172 	[LS_NLA_TYPE_PKEY]		= {.type = NLA_U16},
173 	[LS_NLA_TYPE_QOS_CLASS]		= {.type = NLA_U16},
174 };
175 
176 
177 static void ib_sa_add_one(struct ib_device *device);
178 static void ib_sa_remove_one(struct ib_device *device, void *client_data);
179 
180 static struct ib_client sa_client = {
181 	.name   = "sa",
182 	.add    = ib_sa_add_one,
183 	.remove = ib_sa_remove_one
184 };
185 
186 static DEFINE_SPINLOCK(idr_lock);
187 static DEFINE_IDR(query_idr);
188 
189 static DEFINE_SPINLOCK(tid_lock);
190 static u32 tid;
191 
192 #define PATH_REC_FIELD(field) \
193 	.struct_offset_bytes = offsetof(struct sa_path_rec, field),	\
194 	.struct_size_bytes   = sizeof((struct sa_path_rec *)0)->field,	\
195 	.field_name          = "sa_path_rec:" #field
196 
197 static const struct ib_field path_rec_table[] = {
198 	{ PATH_REC_FIELD(service_id),
199 	  .offset_words = 0,
200 	  .offset_bits  = 0,
201 	  .size_bits    = 64 },
202 	{ PATH_REC_FIELD(dgid),
203 	  .offset_words = 2,
204 	  .offset_bits  = 0,
205 	  .size_bits    = 128 },
206 	{ PATH_REC_FIELD(sgid),
207 	  .offset_words = 6,
208 	  .offset_bits  = 0,
209 	  .size_bits    = 128 },
210 	{ PATH_REC_FIELD(ib.dlid),
211 	  .offset_words = 10,
212 	  .offset_bits  = 0,
213 	  .size_bits    = 16 },
214 	{ PATH_REC_FIELD(ib.slid),
215 	  .offset_words = 10,
216 	  .offset_bits  = 16,
217 	  .size_bits    = 16 },
218 	{ PATH_REC_FIELD(ib.raw_traffic),
219 	  .offset_words = 11,
220 	  .offset_bits  = 0,
221 	  .size_bits    = 1 },
222 	{ RESERVED,
223 	  .offset_words = 11,
224 	  .offset_bits  = 1,
225 	  .size_bits    = 3 },
226 	{ PATH_REC_FIELD(flow_label),
227 	  .offset_words = 11,
228 	  .offset_bits  = 4,
229 	  .size_bits    = 20 },
230 	{ PATH_REC_FIELD(hop_limit),
231 	  .offset_words = 11,
232 	  .offset_bits  = 24,
233 	  .size_bits    = 8 },
234 	{ PATH_REC_FIELD(traffic_class),
235 	  .offset_words = 12,
236 	  .offset_bits  = 0,
237 	  .size_bits    = 8 },
238 	{ PATH_REC_FIELD(reversible),
239 	  .offset_words = 12,
240 	  .offset_bits  = 8,
241 	  .size_bits    = 1 },
242 	{ PATH_REC_FIELD(numb_path),
243 	  .offset_words = 12,
244 	  .offset_bits  = 9,
245 	  .size_bits    = 7 },
246 	{ PATH_REC_FIELD(pkey),
247 	  .offset_words = 12,
248 	  .offset_bits  = 16,
249 	  .size_bits    = 16 },
250 	{ PATH_REC_FIELD(qos_class),
251 	  .offset_words = 13,
252 	  .offset_bits  = 0,
253 	  .size_bits    = 12 },
254 	{ PATH_REC_FIELD(sl),
255 	  .offset_words = 13,
256 	  .offset_bits  = 12,
257 	  .size_bits    = 4 },
258 	{ PATH_REC_FIELD(mtu_selector),
259 	  .offset_words = 13,
260 	  .offset_bits  = 16,
261 	  .size_bits    = 2 },
262 	{ PATH_REC_FIELD(mtu),
263 	  .offset_words = 13,
264 	  .offset_bits  = 18,
265 	  .size_bits    = 6 },
266 	{ PATH_REC_FIELD(rate_selector),
267 	  .offset_words = 13,
268 	  .offset_bits  = 24,
269 	  .size_bits    = 2 },
270 	{ PATH_REC_FIELD(rate),
271 	  .offset_words = 13,
272 	  .offset_bits  = 26,
273 	  .size_bits    = 6 },
274 	{ PATH_REC_FIELD(packet_life_time_selector),
275 	  .offset_words = 14,
276 	  .offset_bits  = 0,
277 	  .size_bits    = 2 },
278 	{ PATH_REC_FIELD(packet_life_time),
279 	  .offset_words = 14,
280 	  .offset_bits  = 2,
281 	  .size_bits    = 6 },
282 	{ PATH_REC_FIELD(preference),
283 	  .offset_words = 14,
284 	  .offset_bits  = 8,
285 	  .size_bits    = 8 },
286 	{ RESERVED,
287 	  .offset_words = 14,
288 	  .offset_bits  = 16,
289 	  .size_bits    = 48 },
290 };
291 
292 #define OPA_PATH_REC_FIELD(field) \
293 	.struct_offset_bytes = \
294 		offsetof(struct sa_path_rec, field), \
295 	.struct_size_bytes   = \
296 		sizeof((struct sa_path_rec *)0)->field,	\
297 	.field_name          = "sa_path_rec:" #field
298 
299 static const struct ib_field opa_path_rec_table[] = {
300 	{ OPA_PATH_REC_FIELD(service_id),
301 	  .offset_words = 0,
302 	  .offset_bits  = 0,
303 	  .size_bits    = 64 },
304 	{ OPA_PATH_REC_FIELD(dgid),
305 	  .offset_words = 2,
306 	  .offset_bits  = 0,
307 	  .size_bits    = 128 },
308 	{ OPA_PATH_REC_FIELD(sgid),
309 	  .offset_words = 6,
310 	  .offset_bits  = 0,
311 	  .size_bits    = 128 },
312 	{ OPA_PATH_REC_FIELD(opa.dlid),
313 	  .offset_words = 10,
314 	  .offset_bits  = 0,
315 	  .size_bits    = 32 },
316 	{ OPA_PATH_REC_FIELD(opa.slid),
317 	  .offset_words = 11,
318 	  .offset_bits  = 0,
319 	  .size_bits    = 32 },
320 	{ OPA_PATH_REC_FIELD(opa.raw_traffic),
321 	  .offset_words = 12,
322 	  .offset_bits  = 0,
323 	  .size_bits    = 1 },
324 	{ RESERVED,
325 	  .offset_words = 12,
326 	  .offset_bits  = 1,
327 	  .size_bits    = 3 },
328 	{ OPA_PATH_REC_FIELD(flow_label),
329 	  .offset_words = 12,
330 	  .offset_bits  = 4,
331 	  .size_bits    = 20 },
332 	{ OPA_PATH_REC_FIELD(hop_limit),
333 	  .offset_words = 12,
334 	  .offset_bits  = 24,
335 	  .size_bits    = 8 },
336 	{ OPA_PATH_REC_FIELD(traffic_class),
337 	  .offset_words = 13,
338 	  .offset_bits  = 0,
339 	  .size_bits    = 8 },
340 	{ OPA_PATH_REC_FIELD(reversible),
341 	  .offset_words = 13,
342 	  .offset_bits  = 8,
343 	  .size_bits    = 1 },
344 	{ OPA_PATH_REC_FIELD(numb_path),
345 	  .offset_words = 13,
346 	  .offset_bits  = 9,
347 	  .size_bits    = 7 },
348 	{ OPA_PATH_REC_FIELD(pkey),
349 	  .offset_words = 13,
350 	  .offset_bits  = 16,
351 	  .size_bits    = 16 },
352 	{ OPA_PATH_REC_FIELD(opa.l2_8B),
353 	  .offset_words = 14,
354 	  .offset_bits  = 0,
355 	  .size_bits    = 1 },
356 	{ OPA_PATH_REC_FIELD(opa.l2_10B),
357 	  .offset_words = 14,
358 	  .offset_bits  = 1,
359 	  .size_bits    = 1 },
360 	{ OPA_PATH_REC_FIELD(opa.l2_9B),
361 	  .offset_words = 14,
362 	  .offset_bits  = 2,
363 	  .size_bits    = 1 },
364 	{ OPA_PATH_REC_FIELD(opa.l2_16B),
365 	  .offset_words = 14,
366 	  .offset_bits  = 3,
367 	  .size_bits    = 1 },
368 	{ RESERVED,
369 	  .offset_words = 14,
370 	  .offset_bits  = 4,
371 	  .size_bits    = 2 },
372 	{ OPA_PATH_REC_FIELD(opa.qos_type),
373 	  .offset_words = 14,
374 	  .offset_bits  = 6,
375 	  .size_bits    = 2 },
376 	{ OPA_PATH_REC_FIELD(opa.qos_priority),
377 	  .offset_words = 14,
378 	  .offset_bits  = 8,
379 	  .size_bits    = 8 },
380 	{ RESERVED,
381 	  .offset_words = 14,
382 	  .offset_bits  = 16,
383 	  .size_bits    = 3 },
384 	{ OPA_PATH_REC_FIELD(sl),
385 	  .offset_words = 14,
386 	  .offset_bits  = 19,
387 	  .size_bits    = 5 },
388 	{ RESERVED,
389 	  .offset_words = 14,
390 	  .offset_bits  = 24,
391 	  .size_bits    = 8 },
392 	{ OPA_PATH_REC_FIELD(mtu_selector),
393 	  .offset_words = 15,
394 	  .offset_bits  = 0,
395 	  .size_bits    = 2 },
396 	{ OPA_PATH_REC_FIELD(mtu),
397 	  .offset_words = 15,
398 	  .offset_bits  = 2,
399 	  .size_bits    = 6 },
400 	{ OPA_PATH_REC_FIELD(rate_selector),
401 	  .offset_words = 15,
402 	  .offset_bits  = 8,
403 	  .size_bits    = 2 },
404 	{ OPA_PATH_REC_FIELD(rate),
405 	  .offset_words = 15,
406 	  .offset_bits  = 10,
407 	  .size_bits    = 6 },
408 	{ OPA_PATH_REC_FIELD(packet_life_time_selector),
409 	  .offset_words = 15,
410 	  .offset_bits  = 16,
411 	  .size_bits    = 2 },
412 	{ OPA_PATH_REC_FIELD(packet_life_time),
413 	  .offset_words = 15,
414 	  .offset_bits  = 18,
415 	  .size_bits    = 6 },
416 	{ OPA_PATH_REC_FIELD(preference),
417 	  .offset_words = 15,
418 	  .offset_bits  = 24,
419 	  .size_bits    = 8 },
420 };
421 
422 #define MCMEMBER_REC_FIELD(field) \
423 	.struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field),	\
424 	.struct_size_bytes   = sizeof ((struct ib_sa_mcmember_rec *) 0)->field,	\
425 	.field_name          = "sa_mcmember_rec:" #field
426 
427 static const struct ib_field mcmember_rec_table[] = {
428 	{ MCMEMBER_REC_FIELD(mgid),
429 	  .offset_words = 0,
430 	  .offset_bits  = 0,
431 	  .size_bits    = 128 },
432 	{ MCMEMBER_REC_FIELD(port_gid),
433 	  .offset_words = 4,
434 	  .offset_bits  = 0,
435 	  .size_bits    = 128 },
436 	{ MCMEMBER_REC_FIELD(qkey),
437 	  .offset_words = 8,
438 	  .offset_bits  = 0,
439 	  .size_bits    = 32 },
440 	{ MCMEMBER_REC_FIELD(mlid),
441 	  .offset_words = 9,
442 	  .offset_bits  = 0,
443 	  .size_bits    = 16 },
444 	{ MCMEMBER_REC_FIELD(mtu_selector),
445 	  .offset_words = 9,
446 	  .offset_bits  = 16,
447 	  .size_bits    = 2 },
448 	{ MCMEMBER_REC_FIELD(mtu),
449 	  .offset_words = 9,
450 	  .offset_bits  = 18,
451 	  .size_bits    = 6 },
452 	{ MCMEMBER_REC_FIELD(traffic_class),
453 	  .offset_words = 9,
454 	  .offset_bits  = 24,
455 	  .size_bits    = 8 },
456 	{ MCMEMBER_REC_FIELD(pkey),
457 	  .offset_words = 10,
458 	  .offset_bits  = 0,
459 	  .size_bits    = 16 },
460 	{ MCMEMBER_REC_FIELD(rate_selector),
461 	  .offset_words = 10,
462 	  .offset_bits  = 16,
463 	  .size_bits    = 2 },
464 	{ MCMEMBER_REC_FIELD(rate),
465 	  .offset_words = 10,
466 	  .offset_bits  = 18,
467 	  .size_bits    = 6 },
468 	{ MCMEMBER_REC_FIELD(packet_life_time_selector),
469 	  .offset_words = 10,
470 	  .offset_bits  = 24,
471 	  .size_bits    = 2 },
472 	{ MCMEMBER_REC_FIELD(packet_life_time),
473 	  .offset_words = 10,
474 	  .offset_bits  = 26,
475 	  .size_bits    = 6 },
476 	{ MCMEMBER_REC_FIELD(sl),
477 	  .offset_words = 11,
478 	  .offset_bits  = 0,
479 	  .size_bits    = 4 },
480 	{ MCMEMBER_REC_FIELD(flow_label),
481 	  .offset_words = 11,
482 	  .offset_bits  = 4,
483 	  .size_bits    = 20 },
484 	{ MCMEMBER_REC_FIELD(hop_limit),
485 	  .offset_words = 11,
486 	  .offset_bits  = 24,
487 	  .size_bits    = 8 },
488 	{ MCMEMBER_REC_FIELD(scope),
489 	  .offset_words = 12,
490 	  .offset_bits  = 0,
491 	  .size_bits    = 4 },
492 	{ MCMEMBER_REC_FIELD(join_state),
493 	  .offset_words = 12,
494 	  .offset_bits  = 4,
495 	  .size_bits    = 4 },
496 	{ MCMEMBER_REC_FIELD(proxy_join),
497 	  .offset_words = 12,
498 	  .offset_bits  = 8,
499 	  .size_bits    = 1 },
500 	{ RESERVED,
501 	  .offset_words = 12,
502 	  .offset_bits  = 9,
503 	  .size_bits    = 23 },
504 };
505 
506 #define SERVICE_REC_FIELD(field) \
507 	.struct_offset_bytes = offsetof(struct ib_sa_service_rec, field),	\
508 	.struct_size_bytes   = sizeof ((struct ib_sa_service_rec *) 0)->field,	\
509 	.field_name          = "sa_service_rec:" #field
510 
511 static const struct ib_field service_rec_table[] = {
512 	{ SERVICE_REC_FIELD(id),
513 	  .offset_words = 0,
514 	  .offset_bits  = 0,
515 	  .size_bits    = 64 },
516 	{ SERVICE_REC_FIELD(gid),
517 	  .offset_words = 2,
518 	  .offset_bits  = 0,
519 	  .size_bits    = 128 },
520 	{ SERVICE_REC_FIELD(pkey),
521 	  .offset_words = 6,
522 	  .offset_bits  = 0,
523 	  .size_bits    = 16 },
524 	{ SERVICE_REC_FIELD(lease),
525 	  .offset_words = 7,
526 	  .offset_bits  = 0,
527 	  .size_bits    = 32 },
528 	{ SERVICE_REC_FIELD(key),
529 	  .offset_words = 8,
530 	  .offset_bits  = 0,
531 	  .size_bits    = 128 },
532 	{ SERVICE_REC_FIELD(name),
533 	  .offset_words = 12,
534 	  .offset_bits  = 0,
535 	  .size_bits    = 64*8 },
536 	{ SERVICE_REC_FIELD(data8),
537 	  .offset_words = 28,
538 	  .offset_bits  = 0,
539 	  .size_bits    = 16*8 },
540 	{ SERVICE_REC_FIELD(data16),
541 	  .offset_words = 32,
542 	  .offset_bits  = 0,
543 	  .size_bits    = 8*16 },
544 	{ SERVICE_REC_FIELD(data32),
545 	  .offset_words = 36,
546 	  .offset_bits  = 0,
547 	  .size_bits    = 4*32 },
548 	{ SERVICE_REC_FIELD(data64),
549 	  .offset_words = 40,
550 	  .offset_bits  = 0,
551 	  .size_bits    = 2*64 },
552 };
553 
554 #define CLASSPORTINFO_REC_FIELD(field) \
555 	.struct_offset_bytes = offsetof(struct ib_class_port_info, field),	\
556 	.struct_size_bytes   = sizeof((struct ib_class_port_info *)0)->field,	\
557 	.field_name          = "ib_class_port_info:" #field
558 
559 static const struct ib_field ib_classport_info_rec_table[] = {
560 	{ CLASSPORTINFO_REC_FIELD(base_version),
561 	  .offset_words = 0,
562 	  .offset_bits  = 0,
563 	  .size_bits    = 8 },
564 	{ CLASSPORTINFO_REC_FIELD(class_version),
565 	  .offset_words = 0,
566 	  .offset_bits  = 8,
567 	  .size_bits    = 8 },
568 	{ CLASSPORTINFO_REC_FIELD(capability_mask),
569 	  .offset_words = 0,
570 	  .offset_bits  = 16,
571 	  .size_bits    = 16 },
572 	{ CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time),
573 	  .offset_words = 1,
574 	  .offset_bits  = 0,
575 	  .size_bits    = 32 },
576 	{ CLASSPORTINFO_REC_FIELD(redirect_gid),
577 	  .offset_words = 2,
578 	  .offset_bits  = 0,
579 	  .size_bits    = 128 },
580 	{ CLASSPORTINFO_REC_FIELD(redirect_tcslfl),
581 	  .offset_words = 6,
582 	  .offset_bits  = 0,
583 	  .size_bits    = 32 },
584 	{ CLASSPORTINFO_REC_FIELD(redirect_lid),
585 	  .offset_words = 7,
586 	  .offset_bits  = 0,
587 	  .size_bits    = 16 },
588 	{ CLASSPORTINFO_REC_FIELD(redirect_pkey),
589 	  .offset_words = 7,
590 	  .offset_bits  = 16,
591 	  .size_bits    = 16 },
592 
593 	{ CLASSPORTINFO_REC_FIELD(redirect_qp),
594 	  .offset_words = 8,
595 	  .offset_bits  = 0,
596 	  .size_bits    = 32 },
597 	{ CLASSPORTINFO_REC_FIELD(redirect_qkey),
598 	  .offset_words = 9,
599 	  .offset_bits  = 0,
600 	  .size_bits    = 32 },
601 
602 	{ CLASSPORTINFO_REC_FIELD(trap_gid),
603 	  .offset_words = 10,
604 	  .offset_bits  = 0,
605 	  .size_bits    = 128 },
606 	{ CLASSPORTINFO_REC_FIELD(trap_tcslfl),
607 	  .offset_words = 14,
608 	  .offset_bits  = 0,
609 	  .size_bits    = 32 },
610 
611 	{ CLASSPORTINFO_REC_FIELD(trap_lid),
612 	  .offset_words = 15,
613 	  .offset_bits  = 0,
614 	  .size_bits    = 16 },
615 	{ CLASSPORTINFO_REC_FIELD(trap_pkey),
616 	  .offset_words = 15,
617 	  .offset_bits  = 16,
618 	  .size_bits    = 16 },
619 
620 	{ CLASSPORTINFO_REC_FIELD(trap_hlqp),
621 	  .offset_words = 16,
622 	  .offset_bits  = 0,
623 	  .size_bits    = 32 },
624 	{ CLASSPORTINFO_REC_FIELD(trap_qkey),
625 	  .offset_words = 17,
626 	  .offset_bits  = 0,
627 	  .size_bits    = 32 },
628 };
629 
630 #define OPA_CLASSPORTINFO_REC_FIELD(field) \
631 	.struct_offset_bytes =\
632 		offsetof(struct opa_class_port_info, field),	\
633 	.struct_size_bytes   = \
634 		sizeof((struct opa_class_port_info *)0)->field,	\
635 	.field_name          = "opa_class_port_info:" #field
636 
637 static const struct ib_field opa_classport_info_rec_table[] = {
638 	{ OPA_CLASSPORTINFO_REC_FIELD(base_version),
639 	  .offset_words = 0,
640 	  .offset_bits  = 0,
641 	  .size_bits    = 8 },
642 	{ OPA_CLASSPORTINFO_REC_FIELD(class_version),
643 	  .offset_words = 0,
644 	  .offset_bits  = 8,
645 	  .size_bits    = 8 },
646 	{ OPA_CLASSPORTINFO_REC_FIELD(cap_mask),
647 	  .offset_words = 0,
648 	  .offset_bits  = 16,
649 	  .size_bits    = 16 },
650 	{ OPA_CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time),
651 	  .offset_words = 1,
652 	  .offset_bits  = 0,
653 	  .size_bits    = 32 },
654 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_gid),
655 	  .offset_words = 2,
656 	  .offset_bits  = 0,
657 	  .size_bits    = 128 },
658 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_tc_fl),
659 	  .offset_words = 6,
660 	  .offset_bits  = 0,
661 	  .size_bits    = 32 },
662 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_lid),
663 	  .offset_words = 7,
664 	  .offset_bits  = 0,
665 	  .size_bits    = 32 },
666 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_sl_qp),
667 	  .offset_words = 8,
668 	  .offset_bits  = 0,
669 	  .size_bits    = 32 },
670 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_qkey),
671 	  .offset_words = 9,
672 	  .offset_bits  = 0,
673 	  .size_bits    = 32 },
674 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_gid),
675 	  .offset_words = 10,
676 	  .offset_bits  = 0,
677 	  .size_bits    = 128 },
678 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_tc_fl),
679 	  .offset_words = 14,
680 	  .offset_bits  = 0,
681 	  .size_bits    = 32 },
682 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_lid),
683 	  .offset_words = 15,
684 	  .offset_bits  = 0,
685 	  .size_bits    = 32 },
686 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_hl_qp),
687 	  .offset_words = 16,
688 	  .offset_bits  = 0,
689 	  .size_bits    = 32 },
690 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_qkey),
691 	  .offset_words = 17,
692 	  .offset_bits  = 0,
693 	  .size_bits    = 32 },
694 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_pkey),
695 	  .offset_words = 18,
696 	  .offset_bits  = 0,
697 	  .size_bits    = 16 },
698 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_pkey),
699 	  .offset_words = 18,
700 	  .offset_bits  = 16,
701 	  .size_bits    = 16 },
702 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_sl_rsvd),
703 	  .offset_words = 19,
704 	  .offset_bits  = 0,
705 	  .size_bits    = 8 },
706 	{ RESERVED,
707 	  .offset_words = 19,
708 	  .offset_bits  = 8,
709 	  .size_bits    = 24 },
710 };
711 
712 #define GUIDINFO_REC_FIELD(field) \
713 	.struct_offset_bytes = offsetof(struct ib_sa_guidinfo_rec, field),	\
714 	.struct_size_bytes   = sizeof((struct ib_sa_guidinfo_rec *) 0)->field,	\
715 	.field_name          = "sa_guidinfo_rec:" #field
716 
717 static const struct ib_field guidinfo_rec_table[] = {
718 	{ GUIDINFO_REC_FIELD(lid),
719 	  .offset_words = 0,
720 	  .offset_bits  = 0,
721 	  .size_bits    = 16 },
722 	{ GUIDINFO_REC_FIELD(block_num),
723 	  .offset_words = 0,
724 	  .offset_bits  = 16,
725 	  .size_bits    = 8 },
726 	{ GUIDINFO_REC_FIELD(res1),
727 	  .offset_words = 0,
728 	  .offset_bits  = 24,
729 	  .size_bits    = 8 },
730 	{ GUIDINFO_REC_FIELD(res2),
731 	  .offset_words = 1,
732 	  .offset_bits  = 0,
733 	  .size_bits    = 32 },
734 	{ GUIDINFO_REC_FIELD(guid_info_list),
735 	  .offset_words = 2,
736 	  .offset_bits  = 0,
737 	  .size_bits    = 512 },
738 };
739 
740 static inline void ib_sa_disable_local_svc(struct ib_sa_query *query)
741 {
742 	query->flags &= ~IB_SA_ENABLE_LOCAL_SERVICE;
743 }
744 
745 static inline int ib_sa_query_cancelled(struct ib_sa_query *query)
746 {
747 	return (query->flags & IB_SA_CANCEL);
748 }
749 
750 static void ib_nl_set_path_rec_attrs(struct sk_buff *skb,
751 				     struct ib_sa_query *query)
752 {
753 	struct sa_path_rec *sa_rec = query->mad_buf->context[1];
754 	struct ib_sa_mad *mad = query->mad_buf->mad;
755 	ib_sa_comp_mask comp_mask = mad->sa_hdr.comp_mask;
756 	u16 val16;
757 	u64 val64;
758 	struct rdma_ls_resolve_header *header;
759 
760 	query->mad_buf->context[1] = NULL;
761 
762 	/* Construct the family header first */
763 	header = skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
764 	memcpy(header->device_name, query->port->agent->device->name,
765 	       LS_DEVICE_NAME_MAX);
766 	header->port_num = query->port->port_num;
767 
768 	if ((comp_mask & IB_SA_PATH_REC_REVERSIBLE) &&
769 	    sa_rec->reversible != 0)
770 		query->path_use = LS_RESOLVE_PATH_USE_GMP;
771 	else
772 		query->path_use = LS_RESOLVE_PATH_USE_UNIDIRECTIONAL;
773 	header->path_use = query->path_use;
774 
775 	/* Now build the attributes */
776 	if (comp_mask & IB_SA_PATH_REC_SERVICE_ID) {
777 		val64 = be64_to_cpu(sa_rec->service_id);
778 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SERVICE_ID,
779 			sizeof(val64), &val64);
780 	}
781 	if (comp_mask & IB_SA_PATH_REC_DGID)
782 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_DGID,
783 			sizeof(sa_rec->dgid), &sa_rec->dgid);
784 	if (comp_mask & IB_SA_PATH_REC_SGID)
785 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SGID,
786 			sizeof(sa_rec->sgid), &sa_rec->sgid);
787 	if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
788 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_TCLASS,
789 			sizeof(sa_rec->traffic_class), &sa_rec->traffic_class);
790 
791 	if (comp_mask & IB_SA_PATH_REC_PKEY) {
792 		val16 = be16_to_cpu(sa_rec->pkey);
793 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_PKEY,
794 			sizeof(val16), &val16);
795 	}
796 	if (comp_mask & IB_SA_PATH_REC_QOS_CLASS) {
797 		val16 = be16_to_cpu(sa_rec->qos_class);
798 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_QOS_CLASS,
799 			sizeof(val16), &val16);
800 	}
801 }
802 
803 static int ib_nl_get_path_rec_attrs_len(ib_sa_comp_mask comp_mask)
804 {
805 	int len = 0;
806 
807 	if (comp_mask & IB_SA_PATH_REC_SERVICE_ID)
808 		len += nla_total_size(sizeof(u64));
809 	if (comp_mask & IB_SA_PATH_REC_DGID)
810 		len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
811 	if (comp_mask & IB_SA_PATH_REC_SGID)
812 		len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
813 	if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
814 		len += nla_total_size(sizeof(u8));
815 	if (comp_mask & IB_SA_PATH_REC_PKEY)
816 		len += nla_total_size(sizeof(u16));
817 	if (comp_mask & IB_SA_PATH_REC_QOS_CLASS)
818 		len += nla_total_size(sizeof(u16));
819 
820 	/*
821 	 * Make sure that at least some of the required comp_mask bits are
822 	 * set.
823 	 */
824 	if (WARN_ON(len == 0))
825 		return len;
826 
827 	/* Add the family header */
828 	len += NLMSG_ALIGN(sizeof(struct rdma_ls_resolve_header));
829 
830 	return len;
831 }
832 
833 static int ib_nl_send_msg(struct ib_sa_query *query, gfp_t gfp_mask)
834 {
835 	struct sk_buff *skb = NULL;
836 	struct nlmsghdr *nlh;
837 	void *data;
838 	int ret = 0;
839 	struct ib_sa_mad *mad;
840 	int len;
841 
842 	mad = query->mad_buf->mad;
843 	len = ib_nl_get_path_rec_attrs_len(mad->sa_hdr.comp_mask);
844 	if (len <= 0)
845 		return -EMSGSIZE;
846 
847 	skb = nlmsg_new(len, gfp_mask);
848 	if (!skb)
849 		return -ENOMEM;
850 
851 	/* Put nlmsg header only for now */
852 	data = ibnl_put_msg(skb, &nlh, query->seq, 0, RDMA_NL_LS,
853 			    RDMA_NL_LS_OP_RESOLVE, NLM_F_REQUEST);
854 	if (!data) {
855 		nlmsg_free(skb);
856 		return -EMSGSIZE;
857 	}
858 
859 	/* Add attributes */
860 	ib_nl_set_path_rec_attrs(skb, query);
861 
862 	/* Repair the nlmsg header length */
863 	nlmsg_end(skb, nlh);
864 
865 	ret = rdma_nl_multicast(skb, RDMA_NL_GROUP_LS, gfp_mask);
866 	if (!ret)
867 		ret = len;
868 	else
869 		ret = 0;
870 
871 	return ret;
872 }
873 
874 static int ib_nl_make_request(struct ib_sa_query *query, gfp_t gfp_mask)
875 {
876 	unsigned long flags;
877 	unsigned long delay;
878 	int ret;
879 
880 	INIT_LIST_HEAD(&query->list);
881 	query->seq = (u32)atomic_inc_return(&ib_nl_sa_request_seq);
882 
883 	/* Put the request on the list first.*/
884 	spin_lock_irqsave(&ib_nl_request_lock, flags);
885 	delay = msecs_to_jiffies(sa_local_svc_timeout_ms);
886 	query->timeout = delay + jiffies;
887 	list_add_tail(&query->list, &ib_nl_request_list);
888 	/* Start the timeout if this is the only request */
889 	if (ib_nl_request_list.next == &query->list)
890 		queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
891 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
892 
893 	ret = ib_nl_send_msg(query, gfp_mask);
894 	if (ret <= 0) {
895 		ret = -EIO;
896 		/* Remove the request */
897 		spin_lock_irqsave(&ib_nl_request_lock, flags);
898 		list_del(&query->list);
899 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
900 	} else {
901 		ret = 0;
902 	}
903 
904 	return ret;
905 }
906 
907 static int ib_nl_cancel_request(struct ib_sa_query *query)
908 {
909 	unsigned long flags;
910 	struct ib_sa_query *wait_query;
911 	int found = 0;
912 
913 	spin_lock_irqsave(&ib_nl_request_lock, flags);
914 	list_for_each_entry(wait_query, &ib_nl_request_list, list) {
915 		/* Let the timeout to take care of the callback */
916 		if (query == wait_query) {
917 			query->flags |= IB_SA_CANCEL;
918 			query->timeout = jiffies;
919 			list_move(&query->list, &ib_nl_request_list);
920 			found = 1;
921 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work, 1);
922 			break;
923 		}
924 	}
925 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
926 
927 	return found;
928 }
929 
930 static void send_handler(struct ib_mad_agent *agent,
931 			 struct ib_mad_send_wc *mad_send_wc);
932 
933 static void ib_nl_process_good_resolve_rsp(struct ib_sa_query *query,
934 					   const struct nlmsghdr *nlh)
935 {
936 	struct ib_mad_send_wc mad_send_wc;
937 	struct ib_sa_mad *mad = NULL;
938 	const struct nlattr *head, *curr;
939 	struct ib_path_rec_data  *rec;
940 	int len, rem;
941 	u32 mask = 0;
942 	int status = -EIO;
943 
944 	if (query->callback) {
945 		head = (const struct nlattr *) nlmsg_data(nlh);
946 		len = nlmsg_len(nlh);
947 		switch (query->path_use) {
948 		case LS_RESOLVE_PATH_USE_UNIDIRECTIONAL:
949 			mask = IB_PATH_PRIMARY | IB_PATH_OUTBOUND;
950 			break;
951 
952 		case LS_RESOLVE_PATH_USE_ALL:
953 		case LS_RESOLVE_PATH_USE_GMP:
954 		default:
955 			mask = IB_PATH_PRIMARY | IB_PATH_GMP |
956 				IB_PATH_BIDIRECTIONAL;
957 			break;
958 		}
959 		nla_for_each_attr(curr, head, len, rem) {
960 			if (curr->nla_type == LS_NLA_TYPE_PATH_RECORD) {
961 				rec = nla_data(curr);
962 				/*
963 				 * Get the first one. In the future, we may
964 				 * need to get up to 6 pathrecords.
965 				 */
966 				if ((rec->flags & mask) == mask) {
967 					mad = query->mad_buf->mad;
968 					mad->mad_hdr.method |=
969 						IB_MGMT_METHOD_RESP;
970 					memcpy(mad->data, rec->path_rec,
971 					       sizeof(rec->path_rec));
972 					status = 0;
973 					break;
974 				}
975 			}
976 		}
977 		query->callback(query, status, mad);
978 	}
979 
980 	mad_send_wc.send_buf = query->mad_buf;
981 	mad_send_wc.status = IB_WC_SUCCESS;
982 	send_handler(query->mad_buf->mad_agent, &mad_send_wc);
983 }
984 
985 static void ib_nl_request_timeout(struct work_struct *work)
986 {
987 	unsigned long flags;
988 	struct ib_sa_query *query;
989 	unsigned long delay;
990 	struct ib_mad_send_wc mad_send_wc;
991 	int ret;
992 
993 	spin_lock_irqsave(&ib_nl_request_lock, flags);
994 	while (!list_empty(&ib_nl_request_list)) {
995 		query = list_entry(ib_nl_request_list.next,
996 				   struct ib_sa_query, list);
997 
998 		if (time_after(query->timeout, jiffies)) {
999 			delay = query->timeout - jiffies;
1000 			if ((long)delay <= 0)
1001 				delay = 1;
1002 			queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
1003 			break;
1004 		}
1005 
1006 		list_del(&query->list);
1007 		ib_sa_disable_local_svc(query);
1008 		/* Hold the lock to protect against query cancellation */
1009 		if (ib_sa_query_cancelled(query))
1010 			ret = -1;
1011 		else
1012 			ret = ib_post_send_mad(query->mad_buf, NULL);
1013 		if (ret) {
1014 			mad_send_wc.send_buf = query->mad_buf;
1015 			mad_send_wc.status = IB_WC_WR_FLUSH_ERR;
1016 			spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1017 			send_handler(query->port->agent, &mad_send_wc);
1018 			spin_lock_irqsave(&ib_nl_request_lock, flags);
1019 		}
1020 	}
1021 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1022 }
1023 
1024 int ib_nl_handle_set_timeout(struct sk_buff *skb,
1025 			     struct nlmsghdr *nlh,
1026 			     struct netlink_ext_ack *extack)
1027 {
1028 	int timeout, delta, abs_delta;
1029 	const struct nlattr *attr;
1030 	unsigned long flags;
1031 	struct ib_sa_query *query;
1032 	long delay = 0;
1033 	struct nlattr *tb[LS_NLA_TYPE_MAX];
1034 	int ret;
1035 
1036 	if (!(nlh->nlmsg_flags & NLM_F_REQUEST) ||
1037 	    !(NETLINK_CB(skb).sk))
1038 		return -EPERM;
1039 
1040 	ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
1041 			nlmsg_len(nlh), ib_nl_policy, NULL);
1042 	attr = (const struct nlattr *)tb[LS_NLA_TYPE_TIMEOUT];
1043 	if (ret || !attr)
1044 		goto settimeout_out;
1045 
1046 	timeout = *(int *) nla_data(attr);
1047 	if (timeout < IB_SA_LOCAL_SVC_TIMEOUT_MIN)
1048 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MIN;
1049 	if (timeout > IB_SA_LOCAL_SVC_TIMEOUT_MAX)
1050 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MAX;
1051 
1052 	delta = timeout - sa_local_svc_timeout_ms;
1053 	if (delta < 0)
1054 		abs_delta = -delta;
1055 	else
1056 		abs_delta = delta;
1057 
1058 	if (delta != 0) {
1059 		spin_lock_irqsave(&ib_nl_request_lock, flags);
1060 		sa_local_svc_timeout_ms = timeout;
1061 		list_for_each_entry(query, &ib_nl_request_list, list) {
1062 			if (delta < 0 && abs_delta > query->timeout)
1063 				query->timeout = 0;
1064 			else
1065 				query->timeout += delta;
1066 
1067 			/* Get the new delay from the first entry */
1068 			if (!delay) {
1069 				delay = query->timeout - jiffies;
1070 				if (delay <= 0)
1071 					delay = 1;
1072 			}
1073 		}
1074 		if (delay)
1075 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work,
1076 					 (unsigned long)delay);
1077 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1078 	}
1079 
1080 settimeout_out:
1081 	return skb->len;
1082 }
1083 
1084 static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh)
1085 {
1086 	struct nlattr *tb[LS_NLA_TYPE_MAX];
1087 	int ret;
1088 
1089 	if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
1090 		return 0;
1091 
1092 	ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
1093 			nlmsg_len(nlh), ib_nl_policy, NULL);
1094 	if (ret)
1095 		return 0;
1096 
1097 	return 1;
1098 }
1099 
1100 int ib_nl_handle_resolve_resp(struct sk_buff *skb,
1101 			      struct nlmsghdr *nlh,
1102 			      struct netlink_ext_ack *extack)
1103 {
1104 	unsigned long flags;
1105 	struct ib_sa_query *query;
1106 	struct ib_mad_send_buf *send_buf;
1107 	struct ib_mad_send_wc mad_send_wc;
1108 	int found = 0;
1109 	int ret;
1110 
1111 	if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
1112 	    !(NETLINK_CB(skb).sk))
1113 		return -EPERM;
1114 
1115 	spin_lock_irqsave(&ib_nl_request_lock, flags);
1116 	list_for_each_entry(query, &ib_nl_request_list, list) {
1117 		/*
1118 		 * If the query is cancelled, let the timeout routine
1119 		 * take care of it.
1120 		 */
1121 		if (nlh->nlmsg_seq == query->seq) {
1122 			found = !ib_sa_query_cancelled(query);
1123 			if (found)
1124 				list_del(&query->list);
1125 			break;
1126 		}
1127 	}
1128 
1129 	if (!found) {
1130 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1131 		goto resp_out;
1132 	}
1133 
1134 	send_buf = query->mad_buf;
1135 
1136 	if (!ib_nl_is_good_resolve_resp(nlh)) {
1137 		/* if the result is a failure, send out the packet via IB */
1138 		ib_sa_disable_local_svc(query);
1139 		ret = ib_post_send_mad(query->mad_buf, NULL);
1140 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1141 		if (ret) {
1142 			mad_send_wc.send_buf = send_buf;
1143 			mad_send_wc.status = IB_WC_GENERAL_ERR;
1144 			send_handler(query->port->agent, &mad_send_wc);
1145 		}
1146 	} else {
1147 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1148 		ib_nl_process_good_resolve_rsp(query, nlh);
1149 	}
1150 
1151 resp_out:
1152 	return skb->len;
1153 }
1154 
1155 static void free_sm_ah(struct kref *kref)
1156 {
1157 	struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
1158 
1159 	rdma_destroy_ah(sm_ah->ah);
1160 	kfree(sm_ah);
1161 }
1162 
1163 void ib_sa_register_client(struct ib_sa_client *client)
1164 {
1165 	atomic_set(&client->users, 1);
1166 	init_completion(&client->comp);
1167 }
1168 EXPORT_SYMBOL(ib_sa_register_client);
1169 
1170 void ib_sa_unregister_client(struct ib_sa_client *client)
1171 {
1172 	ib_sa_client_put(client);
1173 	wait_for_completion(&client->comp);
1174 }
1175 EXPORT_SYMBOL(ib_sa_unregister_client);
1176 
1177 /**
1178  * ib_sa_cancel_query - try to cancel an SA query
1179  * @id:ID of query to cancel
1180  * @query:query pointer to cancel
1181  *
1182  * Try to cancel an SA query.  If the id and query don't match up or
1183  * the query has already completed, nothing is done.  Otherwise the
1184  * query is canceled and will complete with a status of -EINTR.
1185  */
1186 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
1187 {
1188 	unsigned long flags;
1189 	struct ib_mad_agent *agent;
1190 	struct ib_mad_send_buf *mad_buf;
1191 
1192 	spin_lock_irqsave(&idr_lock, flags);
1193 	if (idr_find(&query_idr, id) != query) {
1194 		spin_unlock_irqrestore(&idr_lock, flags);
1195 		return;
1196 	}
1197 	agent = query->port->agent;
1198 	mad_buf = query->mad_buf;
1199 	spin_unlock_irqrestore(&idr_lock, flags);
1200 
1201 	/*
1202 	 * If the query is still on the netlink request list, schedule
1203 	 * it to be cancelled by the timeout routine. Otherwise, it has been
1204 	 * sent to the MAD layer and has to be cancelled from there.
1205 	 */
1206 	if (!ib_nl_cancel_request(query))
1207 		ib_cancel_mad(agent, mad_buf);
1208 }
1209 EXPORT_SYMBOL(ib_sa_cancel_query);
1210 
1211 static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
1212 {
1213 	struct ib_sa_device *sa_dev;
1214 	struct ib_sa_port   *port;
1215 	unsigned long flags;
1216 	u8 src_path_mask;
1217 
1218 	sa_dev = ib_get_client_data(device, &sa_client);
1219 	if (!sa_dev)
1220 		return 0x7f;
1221 
1222 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1223 	spin_lock_irqsave(&port->ah_lock, flags);
1224 	src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
1225 	spin_unlock_irqrestore(&port->ah_lock, flags);
1226 
1227 	return src_path_mask;
1228 }
1229 
1230 static int roce_resolve_route_from_path(struct sa_path_rec *rec,
1231 					const struct ib_gid_attr *attr)
1232 {
1233 	struct rdma_dev_addr dev_addr = {};
1234 	union {
1235 		struct sockaddr     _sockaddr;
1236 		struct sockaddr_in  _sockaddr_in;
1237 		struct sockaddr_in6 _sockaddr_in6;
1238 	} sgid_addr, dgid_addr;
1239 	int ret;
1240 
1241 	if (rec->roce.route_resolved)
1242 		return 0;
1243 	if (!attr || !attr->ndev)
1244 		return -EINVAL;
1245 
1246 	dev_addr.bound_dev_if = attr->ndev->ifindex;
1247 	/* TODO: Use net from the ib_gid_attr once it is added to it,
1248 	 * until than, limit itself to init_net.
1249 	 */
1250 	dev_addr.net = &init_net;
1251 
1252 	rdma_gid2ip(&sgid_addr._sockaddr, &rec->sgid);
1253 	rdma_gid2ip(&dgid_addr._sockaddr, &rec->dgid);
1254 
1255 	/* validate the route */
1256 	ret = rdma_resolve_ip_route(&sgid_addr._sockaddr,
1257 				    &dgid_addr._sockaddr, &dev_addr);
1258 	if (ret)
1259 		return ret;
1260 
1261 	if ((dev_addr.network == RDMA_NETWORK_IPV4 ||
1262 	     dev_addr.network == RDMA_NETWORK_IPV6) &&
1263 	    rec->rec_type != SA_PATH_REC_TYPE_ROCE_V2)
1264 		return -EINVAL;
1265 
1266 	rec->roce.route_resolved = true;
1267 	return 0;
1268 }
1269 
1270 static int init_ah_attr_grh_fields(struct ib_device *device, u8 port_num,
1271 				   struct sa_path_rec *rec,
1272 				   struct rdma_ah_attr *ah_attr,
1273 				   const struct ib_gid_attr *gid_attr)
1274 {
1275 	enum ib_gid_type type = sa_conv_pathrec_to_gid_type(rec);
1276 
1277 	if (!gid_attr) {
1278 		gid_attr = rdma_find_gid_by_port(device, &rec->sgid, type,
1279 						 port_num, NULL);
1280 		if (IS_ERR(gid_attr))
1281 			return PTR_ERR(gid_attr);
1282 	} else
1283 		rdma_hold_gid_attr(gid_attr);
1284 
1285 	rdma_move_grh_sgid_attr(ah_attr, &rec->dgid,
1286 				be32_to_cpu(rec->flow_label),
1287 				rec->hop_limit,	rec->traffic_class,
1288 				gid_attr);
1289 	return 0;
1290 }
1291 
1292 /**
1293  * ib_init_ah_attr_from_path - Initialize address handle attributes based on
1294  *   an SA path record.
1295  * @device: Device associated ah attributes initialization.
1296  * @port_num: Port on the specified device.
1297  * @rec: path record entry to use for ah attributes initialization.
1298  * @ah_attr: address handle attributes to initialization from path record.
1299  * @sgid_attr: SGID attribute to consider during initialization.
1300  *
1301  * When ib_init_ah_attr_from_path() returns success,
1302  * (a) for IB link layer it optionally contains a reference to SGID attribute
1303  * when GRH is present for IB link layer.
1304  * (b) for RoCE link layer it contains a reference to SGID attribute.
1305  * User must invoke rdma_destroy_ah_attr() to release reference to SGID
1306  * attributes which are initialized using ib_init_ah_attr_from_path().
1307  */
1308 int ib_init_ah_attr_from_path(struct ib_device *device, u8 port_num,
1309 			      struct sa_path_rec *rec,
1310 			      struct rdma_ah_attr *ah_attr,
1311 			      const struct ib_gid_attr *gid_attr)
1312 {
1313 	int ret = 0;
1314 
1315 	memset(ah_attr, 0, sizeof(*ah_attr));
1316 	ah_attr->type = rdma_ah_find_type(device, port_num);
1317 	rdma_ah_set_sl(ah_attr, rec->sl);
1318 	rdma_ah_set_port_num(ah_attr, port_num);
1319 	rdma_ah_set_static_rate(ah_attr, rec->rate);
1320 
1321 	if (sa_path_is_roce(rec)) {
1322 		ret = roce_resolve_route_from_path(rec, gid_attr);
1323 		if (ret)
1324 			return ret;
1325 
1326 		memcpy(ah_attr->roce.dmac, sa_path_get_dmac(rec), ETH_ALEN);
1327 	} else {
1328 		rdma_ah_set_dlid(ah_attr, be32_to_cpu(sa_path_get_dlid(rec)));
1329 		if (sa_path_is_opa(rec) &&
1330 		    rdma_ah_get_dlid(ah_attr) == be16_to_cpu(IB_LID_PERMISSIVE))
1331 			rdma_ah_set_make_grd(ah_attr, true);
1332 
1333 		rdma_ah_set_path_bits(ah_attr,
1334 				      be32_to_cpu(sa_path_get_slid(rec)) &
1335 				      get_src_path_mask(device, port_num));
1336 	}
1337 
1338 	if (rec->hop_limit > 0 || sa_path_is_roce(rec))
1339 		ret = init_ah_attr_grh_fields(device, port_num,
1340 					      rec, ah_attr, gid_attr);
1341 	return ret;
1342 }
1343 EXPORT_SYMBOL(ib_init_ah_attr_from_path);
1344 
1345 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
1346 {
1347 	struct rdma_ah_attr ah_attr;
1348 	unsigned long flags;
1349 
1350 	spin_lock_irqsave(&query->port->ah_lock, flags);
1351 	if (!query->port->sm_ah) {
1352 		spin_unlock_irqrestore(&query->port->ah_lock, flags);
1353 		return -EAGAIN;
1354 	}
1355 	kref_get(&query->port->sm_ah->ref);
1356 	query->sm_ah = query->port->sm_ah;
1357 	spin_unlock_irqrestore(&query->port->ah_lock, flags);
1358 
1359 	/*
1360 	 * Always check if sm_ah has valid dlid assigned,
1361 	 * before querying for class port info
1362 	 */
1363 	if ((rdma_query_ah(query->sm_ah->ah, &ah_attr) < 0) ||
1364 	    !rdma_is_valid_unicast_lid(&ah_attr)) {
1365 		kref_put(&query->sm_ah->ref, free_sm_ah);
1366 		return -EAGAIN;
1367 	}
1368 	query->mad_buf = ib_create_send_mad(query->port->agent, 1,
1369 					    query->sm_ah->pkey_index,
1370 					    0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
1371 					    gfp_mask,
1372 					    ((query->flags & IB_SA_QUERY_OPA) ?
1373 					     OPA_MGMT_BASE_VERSION :
1374 					     IB_MGMT_BASE_VERSION));
1375 	if (IS_ERR(query->mad_buf)) {
1376 		kref_put(&query->sm_ah->ref, free_sm_ah);
1377 		return -ENOMEM;
1378 	}
1379 
1380 	query->mad_buf->ah = query->sm_ah->ah;
1381 
1382 	return 0;
1383 }
1384 
1385 static void free_mad(struct ib_sa_query *query)
1386 {
1387 	ib_free_send_mad(query->mad_buf);
1388 	kref_put(&query->sm_ah->ref, free_sm_ah);
1389 }
1390 
1391 static void init_mad(struct ib_sa_query *query, struct ib_mad_agent *agent)
1392 {
1393 	struct ib_sa_mad *mad = query->mad_buf->mad;
1394 	unsigned long flags;
1395 
1396 	memset(mad, 0, sizeof *mad);
1397 
1398 	if (query->flags & IB_SA_QUERY_OPA) {
1399 		mad->mad_hdr.base_version  = OPA_MGMT_BASE_VERSION;
1400 		mad->mad_hdr.class_version = OPA_SA_CLASS_VERSION;
1401 	} else {
1402 		mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
1403 		mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
1404 	}
1405 	mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
1406 	spin_lock_irqsave(&tid_lock, flags);
1407 	mad->mad_hdr.tid           =
1408 		cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
1409 	spin_unlock_irqrestore(&tid_lock, flags);
1410 }
1411 
1412 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
1413 {
1414 	bool preload = gfpflags_allow_blocking(gfp_mask);
1415 	unsigned long flags;
1416 	int ret, id;
1417 
1418 	if (preload)
1419 		idr_preload(gfp_mask);
1420 	spin_lock_irqsave(&idr_lock, flags);
1421 
1422 	id = idr_alloc(&query_idr, query, 0, 0, GFP_NOWAIT);
1423 
1424 	spin_unlock_irqrestore(&idr_lock, flags);
1425 	if (preload)
1426 		idr_preload_end();
1427 	if (id < 0)
1428 		return id;
1429 
1430 	query->mad_buf->timeout_ms  = timeout_ms;
1431 	query->mad_buf->context[0] = query;
1432 	query->id = id;
1433 
1434 	if ((query->flags & IB_SA_ENABLE_LOCAL_SERVICE) &&
1435 	    (!(query->flags & IB_SA_QUERY_OPA))) {
1436 		if (!rdma_nl_chk_listeners(RDMA_NL_GROUP_LS)) {
1437 			if (!ib_nl_make_request(query, gfp_mask))
1438 				return id;
1439 		}
1440 		ib_sa_disable_local_svc(query);
1441 	}
1442 
1443 	ret = ib_post_send_mad(query->mad_buf, NULL);
1444 	if (ret) {
1445 		spin_lock_irqsave(&idr_lock, flags);
1446 		idr_remove(&query_idr, id);
1447 		spin_unlock_irqrestore(&idr_lock, flags);
1448 	}
1449 
1450 	/*
1451 	 * It's not safe to dereference query any more, because the
1452 	 * send may already have completed and freed the query in
1453 	 * another context.
1454 	 */
1455 	return ret ? ret : id;
1456 }
1457 
1458 void ib_sa_unpack_path(void *attribute, struct sa_path_rec *rec)
1459 {
1460 	ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
1461 }
1462 EXPORT_SYMBOL(ib_sa_unpack_path);
1463 
1464 void ib_sa_pack_path(struct sa_path_rec *rec, void *attribute)
1465 {
1466 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
1467 }
1468 EXPORT_SYMBOL(ib_sa_pack_path);
1469 
1470 static bool ib_sa_opa_pathrecord_support(struct ib_sa_client *client,
1471 					 struct ib_device *device,
1472 					 u8 port_num)
1473 {
1474 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1475 	struct ib_sa_port *port;
1476 	unsigned long flags;
1477 	bool ret = false;
1478 
1479 	if (!sa_dev)
1480 		return ret;
1481 
1482 	port = &sa_dev->port[port_num - sa_dev->start_port];
1483 	spin_lock_irqsave(&port->classport_lock, flags);
1484 	if (!port->classport_info.valid)
1485 		goto ret;
1486 
1487 	if (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_OPA)
1488 		ret = opa_get_cpi_capmask2(&port->classport_info.data.opa) &
1489 			OPA_CLASS_PORT_INFO_PR_SUPPORT;
1490 ret:
1491 	spin_unlock_irqrestore(&port->classport_lock, flags);
1492 	return ret;
1493 }
1494 
1495 enum opa_pr_supported {
1496 	PR_NOT_SUPPORTED,
1497 	PR_OPA_SUPPORTED,
1498 	PR_IB_SUPPORTED
1499 };
1500 
1501 /**
1502  * Check if current PR query can be an OPA query.
1503  * Retuns PR_NOT_SUPPORTED if a path record query is not
1504  * possible, PR_OPA_SUPPORTED if an OPA path record query
1505  * is possible and PR_IB_SUPPORTED if an IB path record
1506  * query is possible.
1507  */
1508 static int opa_pr_query_possible(struct ib_sa_client *client,
1509 				 struct ib_device *device,
1510 				 u8 port_num,
1511 				 struct sa_path_rec *rec)
1512 {
1513 	struct ib_port_attr port_attr;
1514 
1515 	if (ib_query_port(device, port_num, &port_attr))
1516 		return PR_NOT_SUPPORTED;
1517 
1518 	if (ib_sa_opa_pathrecord_support(client, device, port_num))
1519 		return PR_OPA_SUPPORTED;
1520 
1521 	if (port_attr.lid >= be16_to_cpu(IB_MULTICAST_LID_BASE))
1522 		return PR_NOT_SUPPORTED;
1523 	else
1524 		return PR_IB_SUPPORTED;
1525 }
1526 
1527 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
1528 				    int status,
1529 				    struct ib_sa_mad *mad)
1530 {
1531 	struct ib_sa_path_query *query =
1532 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1533 
1534 	if (mad) {
1535 		struct sa_path_rec rec;
1536 
1537 		if (sa_query->flags & IB_SA_QUERY_OPA) {
1538 			ib_unpack(opa_path_rec_table,
1539 				  ARRAY_SIZE(opa_path_rec_table),
1540 				  mad->data, &rec);
1541 			rec.rec_type = SA_PATH_REC_TYPE_OPA;
1542 			query->callback(status, &rec, query->context);
1543 		} else {
1544 			ib_unpack(path_rec_table,
1545 				  ARRAY_SIZE(path_rec_table),
1546 				  mad->data, &rec);
1547 			rec.rec_type = SA_PATH_REC_TYPE_IB;
1548 			sa_path_set_dmac_zero(&rec);
1549 
1550 			if (query->conv_pr) {
1551 				struct sa_path_rec opa;
1552 
1553 				memset(&opa, 0, sizeof(struct sa_path_rec));
1554 				sa_convert_path_ib_to_opa(&opa, &rec);
1555 				query->callback(status, &opa, query->context);
1556 			} else {
1557 				query->callback(status, &rec, query->context);
1558 			}
1559 		}
1560 	} else
1561 		query->callback(status, NULL, query->context);
1562 }
1563 
1564 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
1565 {
1566 	struct ib_sa_path_query *query =
1567 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1568 
1569 	kfree(query->conv_pr);
1570 	kfree(query);
1571 }
1572 
1573 /**
1574  * ib_sa_path_rec_get - Start a Path get query
1575  * @client:SA client
1576  * @device:device to send query on
1577  * @port_num: port number to send query on
1578  * @rec:Path Record to send in query
1579  * @comp_mask:component mask to send in query
1580  * @timeout_ms:time to wait for response
1581  * @gfp_mask:GFP mask to use for internal allocations
1582  * @callback:function called when query completes, times out or is
1583  * canceled
1584  * @context:opaque user context passed to callback
1585  * @sa_query:query context, used to cancel query
1586  *
1587  * Send a Path Record Get query to the SA to look up a path.  The
1588  * callback function will be called when the query completes (or
1589  * fails); status is 0 for a successful response, -EINTR if the query
1590  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1591  * occurred sending the query.  The resp parameter of the callback is
1592  * only valid if status is 0.
1593  *
1594  * If the return value of ib_sa_path_rec_get() is negative, it is an
1595  * error code.  Otherwise it is a query ID that can be used to cancel
1596  * the query.
1597  */
1598 int ib_sa_path_rec_get(struct ib_sa_client *client,
1599 		       struct ib_device *device, u8 port_num,
1600 		       struct sa_path_rec *rec,
1601 		       ib_sa_comp_mask comp_mask,
1602 		       int timeout_ms, gfp_t gfp_mask,
1603 		       void (*callback)(int status,
1604 					struct sa_path_rec *resp,
1605 					void *context),
1606 		       void *context,
1607 		       struct ib_sa_query **sa_query)
1608 {
1609 	struct ib_sa_path_query *query;
1610 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1611 	struct ib_sa_port   *port;
1612 	struct ib_mad_agent *agent;
1613 	struct ib_sa_mad *mad;
1614 	enum opa_pr_supported status;
1615 	int ret;
1616 
1617 	if (!sa_dev)
1618 		return -ENODEV;
1619 
1620 	if ((rec->rec_type != SA_PATH_REC_TYPE_IB) &&
1621 	    (rec->rec_type != SA_PATH_REC_TYPE_OPA))
1622 		return -EINVAL;
1623 
1624 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1625 	agent = port->agent;
1626 
1627 	query = kzalloc(sizeof(*query), gfp_mask);
1628 	if (!query)
1629 		return -ENOMEM;
1630 
1631 	query->sa_query.port     = port;
1632 	if (rec->rec_type == SA_PATH_REC_TYPE_OPA) {
1633 		status = opa_pr_query_possible(client, device, port_num, rec);
1634 		if (status == PR_NOT_SUPPORTED) {
1635 			ret = -EINVAL;
1636 			goto err1;
1637 		} else if (status == PR_OPA_SUPPORTED) {
1638 			query->sa_query.flags |= IB_SA_QUERY_OPA;
1639 		} else {
1640 			query->conv_pr =
1641 				kmalloc(sizeof(*query->conv_pr), gfp_mask);
1642 			if (!query->conv_pr) {
1643 				ret = -ENOMEM;
1644 				goto err1;
1645 			}
1646 		}
1647 	}
1648 
1649 	ret = alloc_mad(&query->sa_query, gfp_mask);
1650 	if (ret)
1651 		goto err2;
1652 
1653 	ib_sa_client_get(client);
1654 	query->sa_query.client = client;
1655 	query->callback        = callback;
1656 	query->context         = context;
1657 
1658 	mad = query->sa_query.mad_buf->mad;
1659 	init_mad(&query->sa_query, agent);
1660 
1661 	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
1662 	query->sa_query.release  = ib_sa_path_rec_release;
1663 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
1664 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
1665 	mad->sa_hdr.comp_mask	 = comp_mask;
1666 
1667 	if (query->sa_query.flags & IB_SA_QUERY_OPA) {
1668 		ib_pack(opa_path_rec_table, ARRAY_SIZE(opa_path_rec_table),
1669 			rec, mad->data);
1670 	} else if (query->conv_pr) {
1671 		sa_convert_path_opa_to_ib(query->conv_pr, rec);
1672 		ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1673 			query->conv_pr, mad->data);
1674 	} else {
1675 		ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1676 			rec, mad->data);
1677 	}
1678 
1679 	*sa_query = &query->sa_query;
1680 
1681 	query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE;
1682 	query->sa_query.mad_buf->context[1] = (query->conv_pr) ?
1683 						query->conv_pr : rec;
1684 
1685 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1686 	if (ret < 0)
1687 		goto err3;
1688 
1689 	return ret;
1690 
1691 err3:
1692 	*sa_query = NULL;
1693 	ib_sa_client_put(query->sa_query.client);
1694 	free_mad(&query->sa_query);
1695 err2:
1696 	kfree(query->conv_pr);
1697 err1:
1698 	kfree(query);
1699 	return ret;
1700 }
1701 EXPORT_SYMBOL(ib_sa_path_rec_get);
1702 
1703 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
1704 				    int status,
1705 				    struct ib_sa_mad *mad)
1706 {
1707 	struct ib_sa_service_query *query =
1708 		container_of(sa_query, struct ib_sa_service_query, sa_query);
1709 
1710 	if (mad) {
1711 		struct ib_sa_service_rec rec;
1712 
1713 		ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
1714 			  mad->data, &rec);
1715 		query->callback(status, &rec, query->context);
1716 	} else
1717 		query->callback(status, NULL, query->context);
1718 }
1719 
1720 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
1721 {
1722 	kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
1723 }
1724 
1725 /**
1726  * ib_sa_service_rec_query - Start Service Record operation
1727  * @client:SA client
1728  * @device:device to send request on
1729  * @port_num: port number to send request on
1730  * @method:SA method - should be get, set, or delete
1731  * @rec:Service Record to send in request
1732  * @comp_mask:component mask to send in request
1733  * @timeout_ms:time to wait for response
1734  * @gfp_mask:GFP mask to use for internal allocations
1735  * @callback:function called when request completes, times out or is
1736  * canceled
1737  * @context:opaque user context passed to callback
1738  * @sa_query:request context, used to cancel request
1739  *
1740  * Send a Service Record set/get/delete to the SA to register,
1741  * unregister or query a service record.
1742  * The callback function will be called when the request completes (or
1743  * fails); status is 0 for a successful response, -EINTR if the query
1744  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1745  * occurred sending the query.  The resp parameter of the callback is
1746  * only valid if status is 0.
1747  *
1748  * If the return value of ib_sa_service_rec_query() is negative, it is an
1749  * error code.  Otherwise it is a request ID that can be used to cancel
1750  * the query.
1751  */
1752 int ib_sa_service_rec_query(struct ib_sa_client *client,
1753 			    struct ib_device *device, u8 port_num, u8 method,
1754 			    struct ib_sa_service_rec *rec,
1755 			    ib_sa_comp_mask comp_mask,
1756 			    int timeout_ms, gfp_t gfp_mask,
1757 			    void (*callback)(int status,
1758 					     struct ib_sa_service_rec *resp,
1759 					     void *context),
1760 			    void *context,
1761 			    struct ib_sa_query **sa_query)
1762 {
1763 	struct ib_sa_service_query *query;
1764 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1765 	struct ib_sa_port   *port;
1766 	struct ib_mad_agent *agent;
1767 	struct ib_sa_mad *mad;
1768 	int ret;
1769 
1770 	if (!sa_dev)
1771 		return -ENODEV;
1772 
1773 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1774 	agent = port->agent;
1775 
1776 	if (method != IB_MGMT_METHOD_GET &&
1777 	    method != IB_MGMT_METHOD_SET &&
1778 	    method != IB_SA_METHOD_DELETE)
1779 		return -EINVAL;
1780 
1781 	query = kzalloc(sizeof(*query), gfp_mask);
1782 	if (!query)
1783 		return -ENOMEM;
1784 
1785 	query->sa_query.port     = port;
1786 	ret = alloc_mad(&query->sa_query, gfp_mask);
1787 	if (ret)
1788 		goto err1;
1789 
1790 	ib_sa_client_get(client);
1791 	query->sa_query.client = client;
1792 	query->callback        = callback;
1793 	query->context         = context;
1794 
1795 	mad = query->sa_query.mad_buf->mad;
1796 	init_mad(&query->sa_query, agent);
1797 
1798 	query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
1799 	query->sa_query.release  = ib_sa_service_rec_release;
1800 	mad->mad_hdr.method	 = method;
1801 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
1802 	mad->sa_hdr.comp_mask	 = comp_mask;
1803 
1804 	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
1805 		rec, mad->data);
1806 
1807 	*sa_query = &query->sa_query;
1808 
1809 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1810 	if (ret < 0)
1811 		goto err2;
1812 
1813 	return ret;
1814 
1815 err2:
1816 	*sa_query = NULL;
1817 	ib_sa_client_put(query->sa_query.client);
1818 	free_mad(&query->sa_query);
1819 
1820 err1:
1821 	kfree(query);
1822 	return ret;
1823 }
1824 EXPORT_SYMBOL(ib_sa_service_rec_query);
1825 
1826 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
1827 					int status,
1828 					struct ib_sa_mad *mad)
1829 {
1830 	struct ib_sa_mcmember_query *query =
1831 		container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
1832 
1833 	if (mad) {
1834 		struct ib_sa_mcmember_rec rec;
1835 
1836 		ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1837 			  mad->data, &rec);
1838 		query->callback(status, &rec, query->context);
1839 	} else
1840 		query->callback(status, NULL, query->context);
1841 }
1842 
1843 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
1844 {
1845 	kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
1846 }
1847 
1848 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
1849 			     struct ib_device *device, u8 port_num,
1850 			     u8 method,
1851 			     struct ib_sa_mcmember_rec *rec,
1852 			     ib_sa_comp_mask comp_mask,
1853 			     int timeout_ms, gfp_t gfp_mask,
1854 			     void (*callback)(int status,
1855 					      struct ib_sa_mcmember_rec *resp,
1856 					      void *context),
1857 			     void *context,
1858 			     struct ib_sa_query **sa_query)
1859 {
1860 	struct ib_sa_mcmember_query *query;
1861 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1862 	struct ib_sa_port   *port;
1863 	struct ib_mad_agent *agent;
1864 	struct ib_sa_mad *mad;
1865 	int ret;
1866 
1867 	if (!sa_dev)
1868 		return -ENODEV;
1869 
1870 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1871 	agent = port->agent;
1872 
1873 	query = kzalloc(sizeof(*query), gfp_mask);
1874 	if (!query)
1875 		return -ENOMEM;
1876 
1877 	query->sa_query.port     = port;
1878 	ret = alloc_mad(&query->sa_query, gfp_mask);
1879 	if (ret)
1880 		goto err1;
1881 
1882 	ib_sa_client_get(client);
1883 	query->sa_query.client = client;
1884 	query->callback        = callback;
1885 	query->context         = context;
1886 
1887 	mad = query->sa_query.mad_buf->mad;
1888 	init_mad(&query->sa_query, agent);
1889 
1890 	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
1891 	query->sa_query.release  = ib_sa_mcmember_rec_release;
1892 	mad->mad_hdr.method	 = method;
1893 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
1894 	mad->sa_hdr.comp_mask	 = comp_mask;
1895 
1896 	ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1897 		rec, mad->data);
1898 
1899 	*sa_query = &query->sa_query;
1900 
1901 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1902 	if (ret < 0)
1903 		goto err2;
1904 
1905 	return ret;
1906 
1907 err2:
1908 	*sa_query = NULL;
1909 	ib_sa_client_put(query->sa_query.client);
1910 	free_mad(&query->sa_query);
1911 
1912 err1:
1913 	kfree(query);
1914 	return ret;
1915 }
1916 
1917 /* Support GuidInfoRecord */
1918 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1919 					int status,
1920 					struct ib_sa_mad *mad)
1921 {
1922 	struct ib_sa_guidinfo_query *query =
1923 		container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1924 
1925 	if (mad) {
1926 		struct ib_sa_guidinfo_rec rec;
1927 
1928 		ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1929 			  mad->data, &rec);
1930 		query->callback(status, &rec, query->context);
1931 	} else
1932 		query->callback(status, NULL, query->context);
1933 }
1934 
1935 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1936 {
1937 	kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1938 }
1939 
1940 int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1941 			      struct ib_device *device, u8 port_num,
1942 			      struct ib_sa_guidinfo_rec *rec,
1943 			      ib_sa_comp_mask comp_mask, u8 method,
1944 			      int timeout_ms, gfp_t gfp_mask,
1945 			      void (*callback)(int status,
1946 					       struct ib_sa_guidinfo_rec *resp,
1947 					       void *context),
1948 			      void *context,
1949 			      struct ib_sa_query **sa_query)
1950 {
1951 	struct ib_sa_guidinfo_query *query;
1952 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1953 	struct ib_sa_port *port;
1954 	struct ib_mad_agent *agent;
1955 	struct ib_sa_mad *mad;
1956 	int ret;
1957 
1958 	if (!sa_dev)
1959 		return -ENODEV;
1960 
1961 	if (method != IB_MGMT_METHOD_GET &&
1962 	    method != IB_MGMT_METHOD_SET &&
1963 	    method != IB_SA_METHOD_DELETE) {
1964 		return -EINVAL;
1965 	}
1966 
1967 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1968 	agent = port->agent;
1969 
1970 	query = kzalloc(sizeof(*query), gfp_mask);
1971 	if (!query)
1972 		return -ENOMEM;
1973 
1974 	query->sa_query.port = port;
1975 	ret = alloc_mad(&query->sa_query, gfp_mask);
1976 	if (ret)
1977 		goto err1;
1978 
1979 	ib_sa_client_get(client);
1980 	query->sa_query.client = client;
1981 	query->callback        = callback;
1982 	query->context         = context;
1983 
1984 	mad = query->sa_query.mad_buf->mad;
1985 	init_mad(&query->sa_query, agent);
1986 
1987 	query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
1988 	query->sa_query.release  = ib_sa_guidinfo_rec_release;
1989 
1990 	mad->mad_hdr.method	 = method;
1991 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
1992 	mad->sa_hdr.comp_mask	 = comp_mask;
1993 
1994 	ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
1995 		mad->data);
1996 
1997 	*sa_query = &query->sa_query;
1998 
1999 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
2000 	if (ret < 0)
2001 		goto err2;
2002 
2003 	return ret;
2004 
2005 err2:
2006 	*sa_query = NULL;
2007 	ib_sa_client_put(query->sa_query.client);
2008 	free_mad(&query->sa_query);
2009 
2010 err1:
2011 	kfree(query);
2012 	return ret;
2013 }
2014 EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
2015 
2016 bool ib_sa_sendonly_fullmem_support(struct ib_sa_client *client,
2017 				    struct ib_device *device,
2018 				    u8 port_num)
2019 {
2020 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
2021 	struct ib_sa_port *port;
2022 	bool ret = false;
2023 	unsigned long flags;
2024 
2025 	if (!sa_dev)
2026 		return ret;
2027 
2028 	port  = &sa_dev->port[port_num - sa_dev->start_port];
2029 
2030 	spin_lock_irqsave(&port->classport_lock, flags);
2031 	if ((port->classport_info.valid) &&
2032 	    (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_IB))
2033 		ret = ib_get_cpi_capmask2(&port->classport_info.data.ib)
2034 			& IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT;
2035 	spin_unlock_irqrestore(&port->classport_lock, flags);
2036 	return ret;
2037 }
2038 EXPORT_SYMBOL(ib_sa_sendonly_fullmem_support);
2039 
2040 struct ib_classport_info_context {
2041 	struct completion	done;
2042 	struct ib_sa_query	*sa_query;
2043 };
2044 
2045 static void ib_classportinfo_cb(void *context)
2046 {
2047 	struct ib_classport_info_context *cb_ctx = context;
2048 
2049 	complete(&cb_ctx->done);
2050 }
2051 
2052 static void ib_sa_classport_info_rec_callback(struct ib_sa_query *sa_query,
2053 					      int status,
2054 					      struct ib_sa_mad *mad)
2055 {
2056 	unsigned long flags;
2057 	struct ib_sa_classport_info_query *query =
2058 		container_of(sa_query, struct ib_sa_classport_info_query, sa_query);
2059 	struct ib_sa_classport_cache *info = &sa_query->port->classport_info;
2060 
2061 	if (mad) {
2062 		if (sa_query->flags & IB_SA_QUERY_OPA) {
2063 			struct opa_class_port_info rec;
2064 
2065 			ib_unpack(opa_classport_info_rec_table,
2066 				  ARRAY_SIZE(opa_classport_info_rec_table),
2067 				  mad->data, &rec);
2068 
2069 			spin_lock_irqsave(&sa_query->port->classport_lock,
2070 					  flags);
2071 			if (!status && !info->valid) {
2072 				memcpy(&info->data.opa, &rec,
2073 				       sizeof(info->data.opa));
2074 
2075 				info->valid = true;
2076 				info->data.type = RDMA_CLASS_PORT_INFO_OPA;
2077 			}
2078 			spin_unlock_irqrestore(&sa_query->port->classport_lock,
2079 					       flags);
2080 
2081 		} else {
2082 			struct ib_class_port_info rec;
2083 
2084 			ib_unpack(ib_classport_info_rec_table,
2085 				  ARRAY_SIZE(ib_classport_info_rec_table),
2086 				  mad->data, &rec);
2087 
2088 			spin_lock_irqsave(&sa_query->port->classport_lock,
2089 					  flags);
2090 			if (!status && !info->valid) {
2091 				memcpy(&info->data.ib, &rec,
2092 				       sizeof(info->data.ib));
2093 
2094 				info->valid = true;
2095 				info->data.type = RDMA_CLASS_PORT_INFO_IB;
2096 			}
2097 			spin_unlock_irqrestore(&sa_query->port->classport_lock,
2098 					       flags);
2099 		}
2100 	}
2101 	query->callback(query->context);
2102 }
2103 
2104 static void ib_sa_classport_info_rec_release(struct ib_sa_query *sa_query)
2105 {
2106 	kfree(container_of(sa_query, struct ib_sa_classport_info_query,
2107 			   sa_query));
2108 }
2109 
2110 static int ib_sa_classport_info_rec_query(struct ib_sa_port *port,
2111 					  int timeout_ms,
2112 					  void (*callback)(void *context),
2113 					  void *context,
2114 					  struct ib_sa_query **sa_query)
2115 {
2116 	struct ib_mad_agent *agent;
2117 	struct ib_sa_classport_info_query *query;
2118 	struct ib_sa_mad *mad;
2119 	gfp_t gfp_mask = GFP_KERNEL;
2120 	int ret;
2121 
2122 	agent = port->agent;
2123 
2124 	query = kzalloc(sizeof(*query), gfp_mask);
2125 	if (!query)
2126 		return -ENOMEM;
2127 
2128 	query->sa_query.port = port;
2129 	query->sa_query.flags |= rdma_cap_opa_ah(port->agent->device,
2130 						 port->port_num) ?
2131 				 IB_SA_QUERY_OPA : 0;
2132 	ret = alloc_mad(&query->sa_query, gfp_mask);
2133 	if (ret)
2134 		goto err_free;
2135 
2136 	query->callback = callback;
2137 	query->context = context;
2138 
2139 	mad = query->sa_query.mad_buf->mad;
2140 	init_mad(&query->sa_query, agent);
2141 
2142 	query->sa_query.callback = ib_sa_classport_info_rec_callback;
2143 	query->sa_query.release  = ib_sa_classport_info_rec_release;
2144 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
2145 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_CLASS_PORTINFO);
2146 	mad->sa_hdr.comp_mask	 = 0;
2147 	*sa_query = &query->sa_query;
2148 
2149 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
2150 	if (ret < 0)
2151 		goto err_free_mad;
2152 
2153 	return ret;
2154 
2155 err_free_mad:
2156 	*sa_query = NULL;
2157 	free_mad(&query->sa_query);
2158 
2159 err_free:
2160 	kfree(query);
2161 	return ret;
2162 }
2163 
2164 static void update_ib_cpi(struct work_struct *work)
2165 {
2166 	struct ib_sa_port *port =
2167 		container_of(work, struct ib_sa_port, ib_cpi_work.work);
2168 	struct ib_classport_info_context *cb_context;
2169 	unsigned long flags;
2170 	int ret;
2171 
2172 	/* If the classport info is valid, nothing
2173 	 * to do here.
2174 	 */
2175 	spin_lock_irqsave(&port->classport_lock, flags);
2176 	if (port->classport_info.valid) {
2177 		spin_unlock_irqrestore(&port->classport_lock, flags);
2178 		return;
2179 	}
2180 	spin_unlock_irqrestore(&port->classport_lock, flags);
2181 
2182 	cb_context = kmalloc(sizeof(*cb_context), GFP_KERNEL);
2183 	if (!cb_context)
2184 		goto err_nomem;
2185 
2186 	init_completion(&cb_context->done);
2187 
2188 	ret = ib_sa_classport_info_rec_query(port, 3000,
2189 					     ib_classportinfo_cb, cb_context,
2190 					     &cb_context->sa_query);
2191 	if (ret < 0)
2192 		goto free_cb_err;
2193 	wait_for_completion(&cb_context->done);
2194 free_cb_err:
2195 	kfree(cb_context);
2196 	spin_lock_irqsave(&port->classport_lock, flags);
2197 
2198 	/* If the classport info is still not valid, the query should have
2199 	 * failed for some reason. Retry issuing the query
2200 	 */
2201 	if (!port->classport_info.valid) {
2202 		port->classport_info.retry_cnt++;
2203 		if (port->classport_info.retry_cnt <=
2204 		    IB_SA_CPI_MAX_RETRY_CNT) {
2205 			unsigned long delay =
2206 				msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2207 
2208 			queue_delayed_work(ib_wq, &port->ib_cpi_work, delay);
2209 		}
2210 	}
2211 	spin_unlock_irqrestore(&port->classport_lock, flags);
2212 
2213 err_nomem:
2214 	return;
2215 }
2216 
2217 static void send_handler(struct ib_mad_agent *agent,
2218 			 struct ib_mad_send_wc *mad_send_wc)
2219 {
2220 	struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
2221 	unsigned long flags;
2222 
2223 	if (query->callback)
2224 		switch (mad_send_wc->status) {
2225 		case IB_WC_SUCCESS:
2226 			/* No callback -- already got recv */
2227 			break;
2228 		case IB_WC_RESP_TIMEOUT_ERR:
2229 			query->callback(query, -ETIMEDOUT, NULL);
2230 			break;
2231 		case IB_WC_WR_FLUSH_ERR:
2232 			query->callback(query, -EINTR, NULL);
2233 			break;
2234 		default:
2235 			query->callback(query, -EIO, NULL);
2236 			break;
2237 		}
2238 
2239 	spin_lock_irqsave(&idr_lock, flags);
2240 	idr_remove(&query_idr, query->id);
2241 	spin_unlock_irqrestore(&idr_lock, flags);
2242 
2243 	free_mad(query);
2244 	if (query->client)
2245 		ib_sa_client_put(query->client);
2246 	query->release(query);
2247 }
2248 
2249 static void recv_handler(struct ib_mad_agent *mad_agent,
2250 			 struct ib_mad_send_buf *send_buf,
2251 			 struct ib_mad_recv_wc *mad_recv_wc)
2252 {
2253 	struct ib_sa_query *query;
2254 
2255 	if (!send_buf)
2256 		return;
2257 
2258 	query = send_buf->context[0];
2259 	if (query->callback) {
2260 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
2261 			query->callback(query,
2262 					mad_recv_wc->recv_buf.mad->mad_hdr.status ?
2263 					-EINVAL : 0,
2264 					(struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
2265 		else
2266 			query->callback(query, -EIO, NULL);
2267 	}
2268 
2269 	ib_free_recv_mad(mad_recv_wc);
2270 }
2271 
2272 static void update_sm_ah(struct work_struct *work)
2273 {
2274 	struct ib_sa_port *port =
2275 		container_of(work, struct ib_sa_port, update_task);
2276 	struct ib_sa_sm_ah *new_ah;
2277 	struct ib_port_attr port_attr;
2278 	struct rdma_ah_attr   ah_attr;
2279 	bool grh_required;
2280 
2281 	if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
2282 		pr_warn("Couldn't query port\n");
2283 		return;
2284 	}
2285 
2286 	new_ah = kmalloc(sizeof(*new_ah), GFP_KERNEL);
2287 	if (!new_ah)
2288 		return;
2289 
2290 	kref_init(&new_ah->ref);
2291 	new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
2292 
2293 	new_ah->pkey_index = 0;
2294 	if (ib_find_pkey(port->agent->device, port->port_num,
2295 			 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
2296 		pr_err("Couldn't find index for default PKey\n");
2297 
2298 	memset(&ah_attr, 0, sizeof(ah_attr));
2299 	ah_attr.type = rdma_ah_find_type(port->agent->device,
2300 					 port->port_num);
2301 	rdma_ah_set_dlid(&ah_attr, port_attr.sm_lid);
2302 	rdma_ah_set_sl(&ah_attr, port_attr.sm_sl);
2303 	rdma_ah_set_port_num(&ah_attr, port->port_num);
2304 
2305 	grh_required = rdma_is_grh_required(port->agent->device,
2306 					    port->port_num);
2307 
2308 	/*
2309 	 * The OPA sm_lid of 0xFFFF needs special handling so that it can be
2310 	 * differentiated from a permissive LID of 0xFFFF.  We set the
2311 	 * grh_required flag here so the SA can program the DGID in the
2312 	 * address handle appropriately
2313 	 */
2314 	if (ah_attr.type == RDMA_AH_ATTR_TYPE_OPA &&
2315 	    (grh_required ||
2316 	     port_attr.sm_lid == be16_to_cpu(IB_LID_PERMISSIVE)))
2317 		rdma_ah_set_make_grd(&ah_attr, true);
2318 
2319 	if (ah_attr.type == RDMA_AH_ATTR_TYPE_IB && grh_required) {
2320 		rdma_ah_set_ah_flags(&ah_attr, IB_AH_GRH);
2321 		rdma_ah_set_subnet_prefix(&ah_attr,
2322 					  cpu_to_be64(port_attr.subnet_prefix));
2323 		rdma_ah_set_interface_id(&ah_attr,
2324 					 cpu_to_be64(IB_SA_WELL_KNOWN_GUID));
2325 	}
2326 
2327 	new_ah->ah = rdma_create_ah(port->agent->qp->pd, &ah_attr);
2328 	if (IS_ERR(new_ah->ah)) {
2329 		pr_warn("Couldn't create new SM AH\n");
2330 		kfree(new_ah);
2331 		return;
2332 	}
2333 
2334 	spin_lock_irq(&port->ah_lock);
2335 	if (port->sm_ah)
2336 		kref_put(&port->sm_ah->ref, free_sm_ah);
2337 	port->sm_ah = new_ah;
2338 	spin_unlock_irq(&port->ah_lock);
2339 }
2340 
2341 static void ib_sa_event(struct ib_event_handler *handler,
2342 			struct ib_event *event)
2343 {
2344 	if (event->event == IB_EVENT_PORT_ERR    ||
2345 	    event->event == IB_EVENT_PORT_ACTIVE ||
2346 	    event->event == IB_EVENT_LID_CHANGE  ||
2347 	    event->event == IB_EVENT_PKEY_CHANGE ||
2348 	    event->event == IB_EVENT_SM_CHANGE   ||
2349 	    event->event == IB_EVENT_CLIENT_REREGISTER) {
2350 		unsigned long flags;
2351 		struct ib_sa_device *sa_dev =
2352 			container_of(handler, typeof(*sa_dev), event_handler);
2353 		u8 port_num = event->element.port_num - sa_dev->start_port;
2354 		struct ib_sa_port *port = &sa_dev->port[port_num];
2355 
2356 		if (!rdma_cap_ib_sa(handler->device, port->port_num))
2357 			return;
2358 
2359 		spin_lock_irqsave(&port->ah_lock, flags);
2360 		if (port->sm_ah)
2361 			kref_put(&port->sm_ah->ref, free_sm_ah);
2362 		port->sm_ah = NULL;
2363 		spin_unlock_irqrestore(&port->ah_lock, flags);
2364 
2365 		if (event->event == IB_EVENT_SM_CHANGE ||
2366 		    event->event == IB_EVENT_CLIENT_REREGISTER ||
2367 		    event->event == IB_EVENT_LID_CHANGE ||
2368 		    event->event == IB_EVENT_PORT_ACTIVE) {
2369 			unsigned long delay =
2370 				msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2371 
2372 			spin_lock_irqsave(&port->classport_lock, flags);
2373 			port->classport_info.valid = false;
2374 			port->classport_info.retry_cnt = 0;
2375 			spin_unlock_irqrestore(&port->classport_lock, flags);
2376 			queue_delayed_work(ib_wq,
2377 					   &port->ib_cpi_work, delay);
2378 		}
2379 		queue_work(ib_wq, &sa_dev->port[port_num].update_task);
2380 	}
2381 }
2382 
2383 static void ib_sa_add_one(struct ib_device *device)
2384 {
2385 	struct ib_sa_device *sa_dev;
2386 	int s, e, i;
2387 	int count = 0;
2388 
2389 	s = rdma_start_port(device);
2390 	e = rdma_end_port(device);
2391 
2392 	sa_dev = kzalloc(sizeof *sa_dev +
2393 			 (e - s + 1) * sizeof (struct ib_sa_port),
2394 			 GFP_KERNEL);
2395 	if (!sa_dev)
2396 		return;
2397 
2398 	sa_dev->start_port = s;
2399 	sa_dev->end_port   = e;
2400 
2401 	for (i = 0; i <= e - s; ++i) {
2402 		spin_lock_init(&sa_dev->port[i].ah_lock);
2403 		if (!rdma_cap_ib_sa(device, i + 1))
2404 			continue;
2405 
2406 		sa_dev->port[i].sm_ah    = NULL;
2407 		sa_dev->port[i].port_num = i + s;
2408 
2409 		spin_lock_init(&sa_dev->port[i].classport_lock);
2410 		sa_dev->port[i].classport_info.valid = false;
2411 
2412 		sa_dev->port[i].agent =
2413 			ib_register_mad_agent(device, i + s, IB_QPT_GSI,
2414 					      NULL, 0, send_handler,
2415 					      recv_handler, sa_dev, 0);
2416 		if (IS_ERR(sa_dev->port[i].agent))
2417 			goto err;
2418 
2419 		INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
2420 		INIT_DELAYED_WORK(&sa_dev->port[i].ib_cpi_work,
2421 				  update_ib_cpi);
2422 
2423 		count++;
2424 	}
2425 
2426 	if (!count)
2427 		goto free;
2428 
2429 	ib_set_client_data(device, &sa_client, sa_dev);
2430 
2431 	/*
2432 	 * We register our event handler after everything is set up,
2433 	 * and then update our cached info after the event handler is
2434 	 * registered to avoid any problems if a port changes state
2435 	 * during our initialization.
2436 	 */
2437 
2438 	INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
2439 	ib_register_event_handler(&sa_dev->event_handler);
2440 
2441 	for (i = 0; i <= e - s; ++i) {
2442 		if (rdma_cap_ib_sa(device, i + 1))
2443 			update_sm_ah(&sa_dev->port[i].update_task);
2444 	}
2445 
2446 	return;
2447 
2448 err:
2449 	while (--i >= 0) {
2450 		if (rdma_cap_ib_sa(device, i + 1))
2451 			ib_unregister_mad_agent(sa_dev->port[i].agent);
2452 	}
2453 free:
2454 	kfree(sa_dev);
2455 	return;
2456 }
2457 
2458 static void ib_sa_remove_one(struct ib_device *device, void *client_data)
2459 {
2460 	struct ib_sa_device *sa_dev = client_data;
2461 	int i;
2462 
2463 	if (!sa_dev)
2464 		return;
2465 
2466 	ib_unregister_event_handler(&sa_dev->event_handler);
2467 	flush_workqueue(ib_wq);
2468 
2469 	for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
2470 		if (rdma_cap_ib_sa(device, i + 1)) {
2471 			cancel_delayed_work_sync(&sa_dev->port[i].ib_cpi_work);
2472 			ib_unregister_mad_agent(sa_dev->port[i].agent);
2473 			if (sa_dev->port[i].sm_ah)
2474 				kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
2475 		}
2476 
2477 	}
2478 
2479 	kfree(sa_dev);
2480 }
2481 
2482 int ib_sa_init(void)
2483 {
2484 	int ret;
2485 
2486 	get_random_bytes(&tid, sizeof tid);
2487 
2488 	atomic_set(&ib_nl_sa_request_seq, 0);
2489 
2490 	ret = ib_register_client(&sa_client);
2491 	if (ret) {
2492 		pr_err("Couldn't register ib_sa client\n");
2493 		goto err1;
2494 	}
2495 
2496 	ret = mcast_init();
2497 	if (ret) {
2498 		pr_err("Couldn't initialize multicast handling\n");
2499 		goto err2;
2500 	}
2501 
2502 	ib_nl_wq = alloc_ordered_workqueue("ib_nl_sa_wq", WQ_MEM_RECLAIM);
2503 	if (!ib_nl_wq) {
2504 		ret = -ENOMEM;
2505 		goto err3;
2506 	}
2507 
2508 	INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout);
2509 
2510 	return 0;
2511 
2512 err3:
2513 	mcast_cleanup();
2514 err2:
2515 	ib_unregister_client(&sa_client);
2516 err1:
2517 	return ret;
2518 }
2519 
2520 void ib_sa_cleanup(void)
2521 {
2522 	cancel_delayed_work(&ib_nl_timed_work);
2523 	flush_workqueue(ib_nl_wq);
2524 	destroy_workqueue(ib_nl_wq);
2525 	mcast_cleanup();
2526 	ib_unregister_client(&sa_client);
2527 	idr_destroy(&query_idr);
2528 }
2529